Machine Learning-Accelerated Exploration in Medicinal Pulsed Plasma using Bayesian Optimization

Modern scientific and medical experiments are frequently complex and designed to be multidimensional in nature and tend to generate large amounts of data. However, the exploration of such a large hyperspace to identify the optimal solution is still largely guided by methods based on experience and intuition similar to those applied for simpler one-dimensional experiments. Unfortunately, despite the recent advances in computing and machine learning, the necessary optimization tools to accelerate and guide experiment design are still missing. In this work, we develop a Bayesian optimization methodology and apply it to navigate the large multidimensional space of cold atmospheric pressure plasma-induced bacteria endospore mortality and adaptively design the next experiment that will bring us closer to the optimal solution of maximum endospore mortality. We show that even with limited data, our approach can identify the parameters needed to perform the next experiment that will better optimize experimental design and result. We show that after only a few iterative steps, the algorithm can correctly recover high log kill data points within test scenarios, exemplifying Bayesian optimization’s potential to accelerate the experimental process and decrease cost and time.